基于静-动态特性协同感知的复杂工业过程运行状态评价  

Evaluation of Complex Industrial Process Operating State Based on Static-dynamic Cooperative Perception

作　　者：褚菲 许杨[1,2] 尚超 王福利[4] 高福荣[5] 马小平[1] CHU Fei;XU Yang;SHANG Chao;WANG Fu-Li;GAO Fu-Rong;MA Xiao-Ping(School of Information and Control Engineering,China University of Mining and Technology,Xuzhou 221116;Underground Space Intelligent Control Engineering Research Center of the Ministry of Education,China University of Mining and Technology,Xuzhou 221116;Department of Automation,Tsinghua University,Beijing 100084;College of Information Science and Engineering,Northeastern University,Shenyang 110819;Department of Chemical Engineering,Hong Kong University of Science and Technology,Hong Kong 999077)

机构地区：[1]中国矿业大学信息与控制工程学院,徐州221116 [2]中国矿业大学地下空间智能控制教育部工程研究中心,徐州221116 [3]清华大学自动化系,北京100084 [4]东北大学信息科学与工程学院,沈阳110819 [5]香港科技大学化工系,中国香港999077

出　　处：《自动化学报》2023年第8期1621-1634,共14页Acta Automatica Sinica

基　　金：国家自然科学基金(61973304,62003187,62073060,61873049);江苏省科技计划项目(BK20191339);江苏省六大人才高峰项目(DZXX-045);徐州市科技创新计划项目(KC19055);矿冶过程自动控制技术国家重点实验室开放课题(BGRIMM-KZSKL-2019-10)资助。

摘　　要：针对当前过程监测和运行状态评价方法等对工况信息感知不全面、漏报和误报现象严重等问题,在深入研究工业现场数据静-动态特性协同感知方法的基础上,提出关键性能指标(Key performance indicators, KPI)驱动的慢特征分析(Slow feature analysis, SFA)算法.将关键性能指标信息融入到慢特征分析中,协同感知复杂工业过程的静-动态特性变化,并进一步通过计算潜变量之间的相似度及其一阶差分间的相似度实现对过程稳态和过渡的评价.在此基础上,建立基于静-动态特性协同感知的过程运行状态评价统一框架.针对非优状态,提出基于稀疏学习的非优因素识别方法,实现对非优因素变量的准确识别.最后,通过重介质选煤过程实际生产数据和田纳西·伊斯曼(Tennessee Eastman, TE)过程数据验证了该方法的有效性.Current process monitoring and operation performance evaluation methods suffer from inadequate capturing of process information as well as severe missed and false alarms.By performing in-depth analysis of methods for concurrent monitoring static-dynamic characteristic of industrial data,this paper proposes a key performance indicators(KPI)-driven slow feature analysis(SFA)algorithm.It integrates KPI information into SFA model in order to concurrently capture static-dynamic characteristic changes of complex industrial processes.The similarity between latent variables and that between first-order differences are computed to evaluate the optimality of static and transitional operations.On this basis,a unified framework for process operation performance assessment is established based on an integrated perception of static-dynamic characteristics.A sparse learning-based non-optimal factor identification method is proposed to effectively highlight root-cause variables that cause unsatisfactory performance.The feasibility and effectiveness of the proposed method are validated based on data collected from a realworld dense medium coal preparation process and the Tennessee Eastman(TE)process.

关 键 词：复杂工业过程 运行状态评价 静-动态特性协同 慢特征分析 稀疏学习 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]